Stylizing digital content

Abstract

In implementations of systems for stylizing digital content, a computing device receives digital content having a plurality of content entities. Classified content entities are generated by classifying the plurality of content entities using one or more machine-learning models. A determination is then made regarding correspondence of the classified content entities with visual styles of a digital template. Based on the determined correspondence, the plurality of content entities of the digital content are displayed as having the visual styles, respectively, of the digital template.

What is claimed is: 
     
         1 . A method comprising:
 receiving, by a processing device, digital content having a plurality of content entities;   generating, by the processing device, classified content entities by classifying the plurality of content entities using one or more machine-learning models;   determining, by the processing device, correspondence of the classified content entities with visual styles of a digital template; and   displaying, by the processing device based on the determined correspondence, the plurality of content entities of the digital content as having the visual styles, respectively, of the digital template.   
     
     
         2 . The method as described in  claim 1 , wherein the determining is based on correspondence of the classified content entities with metadata associated with respective said visual styles of the digital template. 
     
     
         3 . The method as described in  claim 1 , wherein the one or more machine-learning models are configured a transformer-over-transformer model. 
     
     
         4 . The method as described in  claim 3 , wherein the transformer-over-transformer model includes:
 a base transformer configured to generate embeddings from the plurality of content entities; and   a top transformer configured to generate the classified content entities based on the embeddings.   
     
     
         5 . The method as described in  claim 1 , further comprising selecting the digital template from a plurality of digital templates. 
     
     
         6 . The method as described in  claim 5 , wherein the selecting is performed responsive to an input received via a user interface. 
     
     
         7 . The method as described in  claim 1 , wherein the one or more machine-learning models are configured to generate embeddings representing context within the content entities and the classified content entities are determined based on context between the content entities. 
     
     
         8 . The method as described in  claim 1 , wherein the content entities include at least one of a sentence or a paragraph and wherein the classified content entities are classified as at least one of a heading, a body, or a list. 
     
     
         9 . The method as described in  claim 1 , wherein the displaying includes additional portions of the digital content having visual styles included in a style package that defines appearance properties for categories of digital content. 
     
     
         10 . A system comprising:
 a base transformer implemented by a processing device using machine learning, the base transformer configured to generate embeddings from content entities included in digital content;   a top transformer implemented by the processing device using machine learning, the top transformer configured to generate classified content entities based on the embeddings; and   a display module implemented by the processing device to display the digital content as having visual styles of a digital template based on correspondence of the classified content entities with respective metadata of the digital template.   
     
     
         11 . The system as described in  claim 10 , wherein the display module is configured to determine correspondence of the classified content entities with metadata associated with respective said visual styles of the digital template. 
     
     
         12 . The system as described in  claim 10 , wherein the display module is configured to select the digital template from a plurality of digital templates. 
     
     
         13 . The system as described in  claim 10 , wherein the embeddings represent context within the content entities and the classified content entities are determined by the display module based on context between the content entities. 
     
     
         14 . The system as described in  claim 10 , wherein the content entities include at least one of a sentence or a paragraph. 
     
     
         15 . The system as described in  claim 10 , wherein the digital template as output by the display module includes additional portions of the digital content having visual styles included in a style package that defines appearance properties for categories of digital content. 
     
     
         16 . The system as described in  claim 10 , wherein the classified content entities are classified as at least one of a heading, a body, or a list. 
     
     
         17 . A method comprising:
 receiving, by a processing device, digital content having a plurality of content entities;   generating, by the processing device, classified content entities by classifying the plurality of content entities using one or more machine-learning models;   receiving, by the processing device, a selection of a digital template from a plurality of digital templates; and   displaying, by the processing device, the plurality of content entities of the digital content as having visual styles of the selected digital template based on correspondence of the classified content entities with respective metadata of the selected digital template.   
     
     
         18 . The method as described in  claim 17 , wherein the selection is performed responsive to an input received via a user interface. 
     
     
         19 . The method as described in  claim 17 , further comprising determining, by the processing device, correspondence of the classified content entities with visual styles of a digital template and wherein the displaying is based on the determining. 
     
     
         20 . The method as described in  claim 17 , wherein the one or more machine-learning models are configured a transformer-over-transformer model including:
 a base transformer configured to generate embeddings from the plurality of content entities; and   a top transformer configured to generate the classified content entities based on the embeddings.